Context window
How much an AI can keep in mind at once, and why long chats lose details.
The context window is everything the AI can see in one go: your instructions, the conversation so far, any files you added, and its own reply. It's big, but it has a limit. When a chat gets very long, older parts can drop out of view.
How it works
- 01Every time you send a message, the app gathers the chat so far plus any files and instructions.
- 02All of it has to fit inside the model's token limit.
- 03If it doesn't fit, older or less important parts get trimmed or summarized, so early details can get lost.
The fridge whiteboard
Think of the whiteboard on your fridge. It fits this week's appointments, the shopping list and the swim-class reminder. Keep adding notes and eventually something gets wiped to make room, usually the oldest one. A context window works the same way. That's why an AI may forget a detail you mentioned 80 messages ago, even though it remembered it perfectly at the start.
Where you'll use it
Fresh chat, fresh board
Planning a birthday party? Start a new chat for it instead of continuing last month's meal-planning thread.
Long documents
Paste in a full contract or client brief and ask for a summary. It all has to fit inside the window.
The recap trick
If the AI loses track in a long chat, give it a short recap of what matters.
Your nap-time experiment
Build the 'chat recap' habit. At the end of a long chat, paste this, then start a fresh one with the summary.
Summarize everything important from our conversation so far in 10 bullet points or fewer: - decisions we made - details about my situation - open to-dos I'll paste this summary into a new chat so we can continue there.